Adjust SDXL v4.0

1533015_training_data.zip

Adjust_SDXL_v4.0.safetensors

<h1 id="v9.0-ultra">v9.0 ULTRA</h1><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">Can we make this lora even cooler? Yes! <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">ULTRA brings a new depth to the lora, further improving:<ul><li>realism</li><li>color grading</li><li>sharpness</li><li>details</li><li>composition</li><li>lighting</li><li>epicness</li></ul>What is more, both Positive and Negative directions make sense, you can see examples in the gallery. Strength:<ul><li>around 1.00 to 2.00</li><li>or -0.50 to -1.00</li></ul>Enjoy, have fun, and share your creations : -)<h1 id="v9.0">v9.0</h1><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">This version is more epic, more rich in details, plus it was trained for both illustrations and realistic photos.Trigger:<ul><li>not needed but...</li><li>the lora was trained with this set of tags and description: '<span style="background-color:rgb(37, 38, 43);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">improved, photo, photorealistic, cinematic, detailed, intricate, sharp, refined, rich textures, fullness,' '<span style="background-color:rgb(37, 38, 43);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">The image is filled with intricate finely detailed textures with every surface object and background element appearing sharp textured and realistic. The composition is rich and full avoiding emptiness and emphasizing both quantity and quality of details'</li><li><span style="background-color:rgb(37, 38, 43);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">these can be used to enhance the effects of the lora alone - but might get great results without it too</li></ul>Strength:<ul><li>usually best around 1.00-1.5 but often 2.00 has the richest details (sometimes thoough less sharp)</li></ul>Enjoy!<h1 id="v8.0">v8.0</h1>This model has mode details and richer backgrounds compared to v7.0.Trigger:<ul><li>No trigger is needed</li><li>if you want to have extra improvements, use these expressions: improved, high_detail, rich_background</li><li>the no trigger prompt will be closer to the original image, if you use these extra tags, it will be a bit further but you will get extra details</li></ul>Strength:<ul><li>between 1.00 to 1.5 seems to be best for SDXL</li><li>between 0.25 to 2.0 seems to be best for ILL, 0.25 works already nicely</li><li>both positive and negative strength works!!! POSITIVE strength adds more details, NEGATIVE strength means less details but more creativitiy - experiment with both</li></ul>Have fun!<h1 id="v7.0">v7.0</h1>This model goes back to the original roots to create adjustments.Background:<ul><li>if you want to have a richer background, just add 'simple background' to the negatives, the lora will amplify its effect</li></ul>SDXL - Strength:<ul><li>both negative and positive works</li><li>you might start -0.30 or +0.30 and go from there, it seems to be the most optimal range but you can go up and low as -2.00 and +2.00</li><li>beyond a certain negative strength the images will have lower and lower level of details and realism which opens the room for creativity - see example in the gallery</li></ul>PONY - Strength:<ul><li>both negative and positive works</li><li>in case of realistic checkpoints, you can go to -1.00 or even further, it will still be able to create good quality images. In case of anime-style checkpoints you can still use the negative side, it will be even less real than normally - this works until -0.5 to -1.</li><li>but on the positive side you can go up to +2.00. Most of the time the real magic happens around 2.00 : -)</li><li>you might go even up to 2.50 or 3.00, in many cases it will be creating still good and even more detailed reasults (see examples in the gallery)</li><li>the level of intricate details above 1.5 strength is simply INSANE!</li></ul>Good exploration, have fun!<h1 id="v6.0">v6.0</h1>The two versions of v6 v6.a and v6.b are both based on the same database but went trhough a little bit different training path. This means that these are two different versions and sometimes you will like v6.a sometimes v6.b. Check the galleries and see for yourself :)This version has lost some of the porn aspects knowledge but learnt to give more details and depth of field, plus bigger changes.Strength:<ul><li>Both negative and postive work</li><li>usually works until -1.00 more or less</li><li>on the postive side it might go up until 1.50-2.00 even</li></ul>Enjoy!<h1 id="v5.0">v5.0</h1>The v5 brings a lot of changes, it is the mixture of the earlier two systems. This model usually does the followings when in the positive or negative range:<ul><li>POSITIVE: focus on sobject, creates more creative changes, adjusts realism, zooms in</li><li>NEGATIVE: focus on the background, zooms out, less realism yet still a lot of creative changes</li></ul>You might want to use it between -1.0 and 1.0 most of the time, you can start around -0.5 and 0.5 to and see where you want to go from there.Make sure to use both directions and check the sample images for both direstions! This is definetely NOT a "positive strength only" lora :-))<h1 id="v4.0">v4.0</h1>The v4 goes back to the roots of the series where the bigger the number the stronger the change and stronger the realism (up to +1.50-2.00) whereas smaller numbers (up to -2.00) means more anime style/less realistic/less detailed images.The lora is not perfect, sometimes gives low-res or pixelated results, be warned (when using with FLUX and SDXL). On the other hand it is in many aspects the best version of the series, so it is definetely worth sharing it.This version has kept many of main category tags from duo versions where the effects are stronger and has some new additions:<ul><li>battle, steampunk, cyberpunk, car, robot, medieval, female model, modern female hairstyle, male model, modern male hairstyle, flaccid penis, uncut penis, cum, cum on face, cum on body, anal sex, oral, sucking, fucking (ie. vaginal sex), vagina, pov, cum in mouth, handjob, asshole, anus, fingering, bukkake, dildo, group blowjob, group sex, licking, rimming, flying machines, floating city/space, (futuristic) car, (futuristic) soldier, robot.</li></ul>The checkpoint will create images with intricate details and detailed background and it excels with Pony.Strength SDXL/FLUX:<ul><li>-2.00 will do strange things but it still works there, not really meaningful above +1.5</li><li>in general you want to work between -1.5 to 1.5 based on how simplistic OR how realistic/detailed you want the image to be.</li><li>see different strength examples in the gallery</li></ul>Strength PONY:<ul><li>the model has a strongest affect on PONY, even the smallest sthrength can have visible adjustments</li><li>0.5 seems to be already a very strong strength, creating 3D-realistic images, 1.00 brings this to the level of very close to photorealism (see the examples)</li><li>teh negative range is useful, it will create more anime-like, less detailed images</li></ul><h1 id="v3.0">v3.0</h1>This version brings a new approach to the series by creating small incremental changes to an image bringing more realism and more details to it. Eventually the changes become very strong while keeping the main idea of the subject and exponentially changing the backgoround.Porposed strength: from 0.00 up to 2.00 - based on your needs. If you want to keep the anime/illustration feel you might use as low as 0.05-0.15, it will still give you good results and create visible changes.<h1 id="dual">Dual</h1><ul><li>Dual is the latest member of the series which is based on near 1000 images</li><li>Dual combines Naughty and the original series into an image improver which knows about porn too</li><li>The name Dual comes from the fact that instead of it working on one dimension (the smaller the number the less realism, the bigger the number the more realism) it is a complex beast which improves some aspects on the positive dimension and improves others on the negative one</li><li>Usually it improves slightly realism/sharpness/details/strength of colors on the positive dimension and diminishes these on the negative one</li><li>On the other hand the negative dimension likes to make much bigger and more creative changes</li><li>Look at the examples and you will see what I mean</li></ul>How to use?<ul><li>my proposal is to start with +1 and -1 both, then explore in the direction you like more</li><li>IMPORTANT: sometimes even 0.05 change in strength can be enough to move the image into another treshold which changes your picture visibly!</li><li>maximum positive strength migh go up to +2.00, minimum is usally around -1.70 and only rarely reaches it -2.00 (you want to explore that end in each case)</li></ul><ul><li>(For info on Naughty, pls scroll down)</li></ul><h1 id="dual-v2.0">Dual v2.0</h1>Dual v2.0 has a detailed tag-system which can further enhance your images. Here are the main tags which were used in the database:<ul><li>battle, steampunk, cyberpunk, car, robot, medieval, female model, modern female hairstyle, male model, modern male hairstyle, flaccid penis, uncut penis, cum, cum on face, cum on body, anal sex, oral, sucking, fucking (ie. vaginal sex), vagina, pov, cum in mouth, handjob, asshole, anus.</li></ul>ORIGINAL DESCRIPTION --&gt;&gt;<h1 id="this-lora-is-basically-a-'make-this-better'-button-:-))))">This lora is basically a 'MAKE THIS BETTER' button : -))))</h1>It improves the images in many ways.It may create changes in the following areas:<ul><li>lighting</li><li>sharpness</li><li>details</li><li>color grading</li><li>composition</li><li>esthetics</li><li>highlight</li><li>saturation</li><li>photorealism</li><li>texture</li><li>emotional expressions/facial expressions</li></ul>Some platforms need you to add the keyword: 'improved' to your prompt.BUT others work well without it (Civitai for example makes great pics without the need to add the keyword). If you add it though, it will add certain new aspects to the picture - explore it :)See how it improves all details and make your picture pop on this side-by-side comparison:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bdfd4e3b-0b6c-4b6f-b69c-e7a974b52a83/width=525/bdfd4e3b-0b6c-4b6f-b69c-e7a974b52a83.jpeg" />It can change not just details but emotions/facial expressions too:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/7c492523-e2cd-4613-b2cb-8723c23d12c9/width=525/7c492523-e2cd-4613-b2cb-8723c23d12c9.jpeg" />You can INCREASE and DECREASE these values, so you can also create a less detailed, less realistic picture if that is your goal. This Aladdin picture was too 3D-ish, the goal was to create somthing similar like in the cartoon. So, here you can use the negative side of the lora to achieve this: <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/850a6647-c16b-4311-8226-c2c3c6b2455b/width=525/850a6647-c16b-4311-8226-c2c3c6b2455b.jpeg" />Proposed strength depends on the picture and your aim, in case of not photorealistic pictures (eg. drawing, manga, anime), as little as 0.01 can make visible improvements, too much though will transform the picture into a more photorealistic image, so you want to experiment with the right strength for your needs.Strength:<ol><li>SDXL: from 0.01 to 1.00 anything can be used based on your goal. In case photos maybe start around 0.5-0.6, in case of drawings/anime/etc. maybe around 0.1 and the go from there.</li><li>FLUX: you can go up to as high as 2.00 for the strongest effect.</li></ol>The SDXL images were created with my base model:Photorealistic All-Purpose, Porn and Gay-Friendly XL Model<a target="_blank" rel="ugc" href="https://civarchive.com/models/659428/photorealistic-all-purpose-porn-and-gay-friendly-xl-model">https://civarchive.com/models/659428/photorealistic-all-purpose-porn-and-gay-friendly-xl-model</a><h1 id="naughty">Naughty</h1>This LORA was trained on 500 HD porn images which means that although has the ability to enhance general pictures, you probably want to use it to enhance porn images.The Naughty version is a complex LORA. How it actually works is yet to be found out. Instead of bigger number meaning bigger resolution and more details, it can be used in both the negative and positive ways and you will have different results. It can greatly enhance your image in both direction but it will focus on different aspects - meaning your pics will be different in one way or the other way. As a starting point for exploration, I would recommend using both 0.75 and -0.75 and see which is more in the direction which you like. Based on my expreiences so far the positive direction usually changes slightly the subject and finetunes it. It also brings it towards a simple background. Whereas negative strength creates more complexity, has a more stronger effect on the picture. But again, this is just a general desctiption, I have not found so far a clear rule of how exactly it works (if you have any theories, do share it with me:)) ).Sometimes even 0,05 change can make a difference in the picture - e.g. no penis on the image at -0.8 but it is there at -0.75:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ff0e6692-492c-42ea-a211-e4e31a766508/width=525/ff0e6692-492c-42ea-a211-e4e31a766508.jpeg" />So, this is really not a straightforaward lora. But from what I have seen so far, it can greatly enable aspects in the photos and enhance them which really makes it worth the trying :)Have fun exploring!

Files with SHA256: 11ad2822b5ae69343058d5bfdcec024d520441d5b2953facfdb6d8683605a70e